I think this is a regression. There was code that removed the state from zk for a core that could not be created. There was a bug in that, in that you only want to do that for new cores and not existing cores (think cores that existed on startup).
Someone commented out that code while working on an unrelated issue I think. But that was not the correct fix (though its better than the previous dangerous behavior) - it really just needs to be more selective. Can you file a JIRA issue? - Mark http://about.me/markrmiller On Feb 21, 2014, at 12:25 AM, Chia-Chun Shih <chiachun.s...@gmail.com> wrote: > Hi all, > > This is my test procedure: > > 1. start a Zookeeper ensemble and a SolrCloud node > 2. stop Zookeeper ensemble > 3. start Zookeeper ensemble > 4. fail to create a collection (with 1 shard and 1 replica) because of > timeout > 5. restart the SolrCloud node > 6. fail to create a collection with the same name in step 4 because the > collection already exists. But the collection doesn't assign to any > SolrCloud node. > > I am using Solr 4.6.1 and Zookeeper 3.4.5 > > Thanks, > Chia-Chun
